|
1
|
|
|
"""!
|
|
2
|
|
|
|
|
3
|
|
|
@brief Cluster analysis algorithm: K-Medoids (PAM - Partitioning Around Medoids).
|
|
4
|
|
|
@details Based on book description:
|
|
5
|
|
|
- A.K. Jain, R.C Dubes, Algorithms for Clustering Data. 1988.
|
|
6
|
|
|
- L. Kaufman, P.J. Rousseeuw, Finding Groups in Data: an Introduction to Cluster Analysis. 1990.
|
|
7
|
|
|
|
|
8
|
|
|
@authors Andrei Novikov ([email protected])
|
|
9
|
|
|
@date 2014-2016
|
|
10
|
|
|
@copyright GNU Public License
|
|
11
|
|
|
|
|
12
|
|
|
@cond GNU_PUBLIC_LICENSE
|
|
13
|
|
|
PyClustering is free software: you can redistribute it and/or modify
|
|
14
|
|
|
it under the terms of the GNU General Public License as published by
|
|
15
|
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
16
|
|
|
(at your option) any later version.
|
|
17
|
|
|
|
|
18
|
|
|
PyClustering is distributed in the hope that it will be useful,
|
|
19
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
20
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
21
|
|
|
GNU General Public License for more details.
|
|
22
|
|
|
|
|
23
|
|
|
You should have received a copy of the GNU General Public License
|
|
24
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
25
|
|
|
@endcond
|
|
26
|
|
|
|
|
27
|
|
|
"""
|
|
28
|
|
|
|
|
29
|
|
|
|
|
30
|
|
|
from pyclustering.utils import euclidean_distance_sqrt, median;
|
|
31
|
|
|
|
|
32
|
|
|
import pyclustering.core.kmedoids_wrapper as wrapper;
|
|
33
|
|
|
|
|
34
|
|
|
|
|
35
|
|
|
class kmedoids:
|
|
36
|
|
|
"""!
|
|
37
|
|
|
@brief Class represents clustering algorithm K-Medoids (another one title is PAM - Parti).
|
|
38
|
|
|
@details The algorithm is less sensitive to outliers tham K-Means. The principle difference between K-Medoids and K-Medians is that
|
|
39
|
|
|
K-Medoids uses existed points from input data space as medoids, but median in K-Medians can be unreal object (not from
|
|
40
|
|
|
input data space).
|
|
41
|
|
|
|
|
42
|
|
|
Example:
|
|
43
|
|
|
@code
|
|
44
|
|
|
# load list of points for cluster analysis
|
|
45
|
|
|
sample = read_sample(path);
|
|
46
|
|
|
|
|
47
|
|
|
# create instance of K-Medoids algorithm
|
|
48
|
|
|
kmedians_instance = kmedians(sample, [1, 10]);
|
|
49
|
|
|
|
|
50
|
|
|
# run cluster analysis and obtain results
|
|
51
|
|
|
kmedians_instance.process();
|
|
52
|
|
|
kmedians_instance.get_clusters();
|
|
53
|
|
|
@endcode
|
|
54
|
|
|
|
|
55
|
|
|
"""
|
|
56
|
|
|
|
|
57
|
|
|
|
|
58
|
|
|
def __init__(self, data, initial_index_medoids, tolerance = 0.25, ccore = False):
|
|
59
|
|
|
"""!
|
|
60
|
|
|
@brief Constructor of clustering algorithm K-Medoids.
|
|
61
|
|
|
|
|
62
|
|
|
@param[in] data (list): Input data that is presented as list of points (objects), each point should be represented by list or tuple.
|
|
63
|
|
|
@param[in] initial_index_medoids (list): Indexes of intial medoids (indexes of points in input data).
|
|
64
|
|
|
@param[in] tolerance (double): Stop condition: if maximum value of distance change of medoids of clusters is less than tolerance than algorithm will stop processing.
|
|
65
|
|
|
@param[in] ccore (bool): If specified than CCORE library (C++ pyclustering library) is used for clustering instead of Python code.
|
|
66
|
|
|
|
|
67
|
|
|
"""
|
|
68
|
|
|
self.__pointer_data = data;
|
|
69
|
|
|
self.__clusters = [];
|
|
70
|
|
|
self.__medoids = initial_index_medoids;
|
|
71
|
|
|
self.__tolerance = tolerance;
|
|
72
|
|
|
self.__ccore = ccore;
|
|
73
|
|
|
|
|
74
|
|
|
|
|
75
|
|
View Code Duplication |
def process(self):
|
|
|
|
|
|
|
76
|
|
|
"""!
|
|
77
|
|
|
@brief Performs cluster analysis in line with rules of K-Medoids algorithm.
|
|
78
|
|
|
|
|
79
|
|
|
@remark Results of clustering can be obtained using corresponding get methods.
|
|
80
|
|
|
|
|
81
|
|
|
@see get_clusters()
|
|
82
|
|
|
@see get_medoids()
|
|
83
|
|
|
|
|
84
|
|
|
"""
|
|
85
|
|
|
|
|
86
|
|
|
if (self.__ccore is True):
|
|
87
|
|
|
self.__clusters = wrapper.kmedoids(self.__pointer_data, self.__medoids, self.__tolerance);
|
|
88
|
|
|
self.__medoids = self.__update_medoids();
|
|
89
|
|
|
|
|
90
|
|
|
else:
|
|
91
|
|
|
self.__medoids = [ self.__pointer_data[medoid_index] for medoid_index in self.__medoids ];
|
|
92
|
|
|
|
|
93
|
|
|
changes = float('inf');
|
|
94
|
|
|
|
|
95
|
|
|
stop_condition = self.__tolerance * self.__tolerance; # Fast solution
|
|
96
|
|
|
#stop_condition = self.__tolerance; # Slow solution
|
|
97
|
|
|
|
|
98
|
|
|
while (changes > stop_condition):
|
|
99
|
|
|
self.__clusters = self.__update_clusters();
|
|
100
|
|
|
updated_medoids = self.__update_medoids(); # changes should be calculated before asignment
|
|
101
|
|
|
|
|
102
|
|
|
changes = max([euclidean_distance_sqrt(self.__medoids[index], updated_medoids[index]) for index in range(len(updated_medoids))]); # Fast solution
|
|
103
|
|
|
|
|
104
|
|
|
self.__medoids = updated_medoids;
|
|
105
|
|
|
|
|
106
|
|
|
|
|
107
|
|
|
def get_clusters(self):
|
|
108
|
|
|
"""!
|
|
109
|
|
|
@brief Returns list of allocated clusters, each cluster contains indexes of objects in list of data.
|
|
110
|
|
|
|
|
111
|
|
|
@see process()
|
|
112
|
|
|
@see get_medoids()
|
|
113
|
|
|
|
|
114
|
|
|
"""
|
|
115
|
|
|
|
|
116
|
|
|
return self.__clusters;
|
|
117
|
|
|
|
|
118
|
|
|
|
|
119
|
|
|
def get_medoids(self):
|
|
120
|
|
|
"""!
|
|
121
|
|
|
@brief Returns list of medoids of allocated clusters.
|
|
122
|
|
|
|
|
123
|
|
View Code Duplication |
@see process()
|
|
|
|
|
|
|
124
|
|
|
@see get_clusters()
|
|
125
|
|
|
|
|
126
|
|
|
"""
|
|
127
|
|
|
|
|
128
|
|
|
return self.__medoids;
|
|
129
|
|
|
|
|
130
|
|
|
|
|
131
|
|
|
def __update_clusters(self):
|
|
132
|
|
|
"""!
|
|
133
|
|
|
@brief Calculate distance to each point from the each cluster.
|
|
134
|
|
|
@details Nearest points are captured by according clusters and as a result clusters are updated.
|
|
135
|
|
|
|
|
136
|
|
|
@return (list) updated clusters as list of clusters where each cluster contains indexes of objects from data.
|
|
137
|
|
|
|
|
138
|
|
|
"""
|
|
139
|
|
|
|
|
140
|
|
|
clusters = [[] for i in range(len(self.__medoids))];
|
|
141
|
|
|
for index_point in range(len(self.__pointer_data)):
|
|
142
|
|
|
index_optim = -1;
|
|
143
|
|
|
dist_optim = 0.0;
|
|
144
|
|
|
|
|
145
|
|
|
for index in range(len(self.__medoids)):
|
|
146
|
|
|
dist = euclidean_distance_sqrt(self.__pointer_data[index_point], self.__medoids[index]);
|
|
147
|
|
|
|
|
148
|
|
|
if ( (dist < dist_optim) or (index is 0)):
|
|
149
|
|
|
index_optim = index;
|
|
150
|
|
|
dist_optim = dist;
|
|
151
|
|
|
|
|
152
|
|
|
clusters[index_optim].append(index_point);
|
|
153
|
|
|
|
|
154
|
|
|
# If cluster is not able to capture object it should be removed
|
|
155
|
|
|
clusters = [cluster for cluster in clusters if len(cluster) > 0];
|
|
156
|
|
|
|
|
157
|
|
|
return clusters;
|
|
158
|
|
|
|
|
159
|
|
|
|
|
160
|
|
|
def __update_medoids(self):
|
|
161
|
|
|
"""!
|
|
162
|
|
|
@brief Find medoids of clusters in line with contained objects.
|
|
163
|
|
|
|
|
164
|
|
|
@return (list) list of medoids for current number of clusters.
|
|
165
|
|
|
|
|
166
|
|
|
"""
|
|
167
|
|
|
|
|
168
|
|
|
medoids = [[] for i in range(len(self.__clusters))];
|
|
169
|
|
|
|
|
170
|
|
|
for index in range(len(self.__clusters)):
|
|
171
|
|
|
medoid_index = median(self.__pointer_data, self.__clusters[index]);
|
|
172
|
|
|
medoids[index] = self.__pointer_data[medoid_index];
|
|
173
|
|
|
|
|
174
|
|
|
return medoids; |
annoviko /
pyclustering
Loading history...